Re: [Axiom-developer] sbcl port

[Waldek Hebisch]

Gregory Vanuxem wrote:
> Le lundi 14 mai 2007 ? 01:08 +0200, Waldek Hebisch a écrit
> 
> [...]
> 
> > 3) The reclos test do not finish, it looks that 
> > 
> > relativeApprox(squareDiff8,10**(-3))::Float
> > 
> > line goes into infinite loop.
> 
> Does
> 
> )lis (gcd 8 most-negative-fixnum)
> 
> return the gcd ?
> 
> If not it's a bug in SBCL (x86_64). I reported it some
> times ago and didn't receive any response. Maybe another
> report with an example will help (I patch sb-kernel::two-arg-gcd each
> time I upgrade SBCL).
> 

Thanks for info.  Yes, as of 1.0.4 declarations in two-arg-gcd are
blatantly wrong.  I must say that it is strange that the code works
at all (C compiler would immediatly punish them with non-working
code).  I also looked at SBCL CVS repository and the wrong code is
still there...

I remeber that you wrote about SBCL bug, but I supected that you
mean that SBCL miscompiles some parts of AXIOM -- build log suggests
that some routines are miscompiled. 

After patching two-arg-gcd it seems that my private version passes
all tests (modulo missing writablep function and fake sockets).


-- 
                              Waldek Hebisch
address@hidden